Advances in biological and medical technologies have been providing us explosive volumes of biological and physiological data, such as medical images, electroencephalography, genomic and protein sequences. Learning from these data facilitates the understanding of human health and disease. Developed from artificial neural networks, deep learning-based algorithms show great promise in extracting features and learning patterns from complex data. The aim of this paper is to provide an overview of deep learning techniques and some of the state-of-the-art applications in the biomedical field. We first introduce the development of artificial neural network and deep learning. We then describe two main components of deep learning, i.e., deep learning architectures and model optimization. Subsequently, some examples are demonstrated for deep learning applications, including medical image classification, genomic sequence analysis, as well as protein structure classification and prediction. Finally, we offer our perspectives for the future directions in the field of deep learning.

Chemical passivation via functional additives plays a critical role in achieving high performance perovskite light‐emitting diodes (PeLEDs). Here, perovskite composite films for high performance PeLEDs by using zwitterion 3‐aminopropanesulfonic acid (APS) as the additive are developed. The sulfonic group of APS can simultaneously passivate deep and shallow level defects in perovskites via coordinate and hydrogen bonding, which leads to suppressed non‐radiative recombination and ion migration in the perovskite composite films. Based on this, PeLEDs with a peak external quantum efficiency of 19.2% and a half‐lifetime of 43 h at a constant current density of 100 mA cm−2 are obtained, representing one of the most stable and efficient PeLEDs under high current densities.

BackgroundC4 photosynthesis evolved from C3 photosynthesis and has higher light, water, and nitrogen use efficiencies. Several C4 photosynthesis genes show cell-specific expression patterns, which are required for these high resource-use efficiencies. However, the mechanisms underlying the evolution of cis-regulatory elements that control these cell-specific expression patterns remain elusive.ResultsIn the present study, we tested the hypothesis that the cis-regulatory motifs related to C4 photosynthesis genes were recruited from non-photosynthetic genes and further examined potential mechanisms facilitating this recruitment. We examined 65 predicted bundle sheath cell-specific motifs, 17 experimentally validated cell-specific cis-regulatory elements, and 1,034 motifs derived from gene regulatory networks. Approximately 7, 5, and 1,000 of these three categories of motifs, respectively, were apparently recruited during the evolution of C4 photosynthesis. In addition, we checked 1) the distance between the acceptors and the donors of potentially recruited motifs in a chromosome, and 2) whether the potentially recruited motifs reside within the overlapping region of transposable elements and the promoter of donor genes. The results showed that 7, 4, and 658 of the potentially recruited motifs might have moved via the transposable elements. Furthermore, the potentially recruited motifs showed higher binding affinity to transcription factors compared to randomly generated sequences of the same length as the motifs.ConclusionsThis study provides molecular evidence supporting the hypothesis that transposon-driven recruitment of pre-existing cis-regulatory elements from non-photosynthetic genes into photosynthetic genes plays an important role during C4 evolution. The findings of the present study coincide with the observed repetitive emergence of C4 during evolution.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-016-2519-3) contains supplementary material, which is available to authorized users.

Additive engineering with organic molecules is of critical importance for achieving high‐performance perovskite optoelectronic devices. However, experimentally finding suitable additives is costly and time consuming, while conventional machine learning (ML) is difficult to predict accurately due to the limited experimental data available in this relatively new field. Here, we demonstrate a deep learning method that can predict the effectiveness of additives in perovskite light‐emitting diodes (PeLEDs) with a high accuracy up to 96 % by using a small dataset of 132 molecules. This model can maximize the information of the molecules and significantly mitigate the duplicated problem that usually happened with previous models in ML for molecular screening. Very high efficiency PeLEDs with a peak external quantum efficiency up to 22.7 % can be achieved by using the predicated additive. Our work opens a new avenue for further boosting the performance of perovskite optoelectronic devices.